Simultaneous truth and performance level estimation through fusion of probabilistic segmentations

Abstract

Recent research has demonstrated that improved image segmentation can be achieved by multiple template fusion utilizing both label and intensity information. However, intensity weighted fusion approaches use local intensity similarity as a surrogate measure of local template quality for predicting target segmentation and do not seek to characterize template performance. This limits both the usefulness and accuracy of these techniques. Our work here was motivated by the observation that the local intensity similarity is a poor surrogate measure for direct comparison of the template image with the true image target segmentation. Although the true image target segmentation is not available, a high quality estimate can be inferred, and this in turn allows a principled estimate to be made of the local quality of each template at contributing to the target segmentation. We developed a fusion algorithm that uses probabilistic segmentations of the target image to simultaneously infer a reference standard segmentation of the target image and the local quality of each probabilistic segmentation. The concept of comparing templates to a hidden reference standard segmentation enables accurate assessments of the contribution of each template to inferring the target image segmentation to be made, and in practice leads to excellent target image segmentation. We have used the new algorithm for the multiple-template-based segmentation and parcellation of magnetic resonance images of the brain. Intensity and label map images of each one of the aligned templates are used to train a local Gaussian mixture model based classifier. Then, each classifier is used to compute the probabilistic segmentations of the target image. Finally, the generated probabilistic segmentations are fused together using the new fusion algorithm to obtain the segmentation of the target image. We evaluated our method in comparison to other state-of-the-art segmentation methods. We demonstrated that our new fusion algorithm has higher segmentation performance than these methods.

Fig. 1. Illustration of Synthetic Data Segmentation Results

Comparison of segmentation results generated by PSTAPLE (h), and STAPLE (i) using three samples from (a,d), three samples from (b,e), and two samples from (c,f). The target image is shown in (g) and the corresponding segmentation is shown in (h). It can be seen that using intensity information PSTAPLE has perfect segmentation accuracy. However, STAPLE could not find the correct answer because there is no way to distinguish between the template segmentations.

Fig. 2. Quantitative Comparison of the IBSR Multi-Atlas Segmentation Results

Comparison of generalized Dice coefficient of M₁₂: Local MAP STAPLE and M₁₃: Local MAP PSTAPLE for 34 structures in 18 IBSR datasets. It indicates that Local MAP PSTAPLE is superior to the Local STAPLE. The horizontal axis represents each subject.

Fig. 3. Quantitative Comparison of the IBSR Multi-Atlas Segmentation Results

Comparison of generalized Dice coefficient of M₁₃: Local MAP PSTAPLE, M₁: method of Artaechevarria et al. [1], M₂: method of Sabuncu et al. [2], M₇: Multi-STEPS [18], M₈: NLS [19], and M₉: PICSL-MALF [25] for 34 structures in 18 IBSR datasets. It indicates that Local MAP PSTAPLE is superior to the other methods. The horizontal axis represents each subject.

Fig. 4. Illustration of IBSR Multi-Atlas Segmentation Results

Comparison of segmentation results generated by M₁₃: Local MAP PSTAPLE (f), M1: method of [1] (a), M₂: method of [2] (b), M₇: Multi-STEPS [18] (c), M₈: NLS [19] (d), M₉: PICSL-MALF [25] (e), and expert manual segmentation (g) in an axial image of a representative IBSR dataset (h). Circles show the regions that Local MAP PSTAPLE clearly outperforms other methods.

Fig. 5. Illustration of NMM Multi-Atlas Segmentation Results

Comparison of segmentation results generated by M₁₃: Local MAP PSTAPLE (f), M₁: method of [1] (a), M₂: method of [2] (b), M₇: Multi-STEPS [18] (c), M₈: NLS [19] (d), M₉: PICSL-MALF [25] (e) , and expert manual segmentation (g) in coronal images of representative NMM datasets (h). Circles show the regions that Local MAP PSTAPLE clearly outperforms other methods.

Fig. 6. Quantitative Comparison of the NMM Multi-Atlas Segmentation Results

Comparison of generalized Dice coefficient of M₁₃: Local MAP PSTAPLE and methods of M₁ [1], M₂ [2], M₇ [18], M₈ [19], and M₉ for 53 structures in 15 NMM datasets. It indicates that Local MAP PSTAPLE is superior to the other methods. The horizontal axis represents each subject.